Re: [PATCH] opensm: Reduce heap consumption by multicast routing tables (MFTs)

[Sasha Khapyorsky <sashak-smomgflXvOZWk0Htik3J/w@public.gmane.org>]

Hi Hal,

On 07:14 Wed 14 Oct     , Hal Rosenstock wrote:
> 
> Heap memory consumption by the unicast and multicast routing tables can be
> reduced.
> 
> This patch is analagous to the previous patch doing this for the unicast
> routing tables (LFTs).
> 
> Using valgrind --tool=massif (for heap profiling), there are couple of place
> ->38.75% (11,206,656B) 0x43267E: osm_switch_new (osm_switch.c:134)
> ->12.89% (3,728,256B) 0x40F8C9: osm_mcast_tbl_init (osm_mcast_tbl.c:96)
> 
> osm_mcast_tbl_init (osm_mcast_tbl.c:96):
>         p_tbl->p_mask_tbl = malloc(p_tbl->num_entries *
>                                    (IB_MCAST_POSITION_MAX +
>                                     1) * IB_MCAST_MASK_SIZE / 8);
> 
> num_entries above is set based on the switch's MulticastFDBCap
> (indicated in it's SM class SwitchInfo attribute).
> 
> The MFT is now allocated in chunks. If there is a MLID offset that exceeds
> the current MFT size, the MFT is reallocated with an increased size.
> 
> A new config option specifies the MFT allocation chunk (in units of 64 entries)
> and defaults to 1K entries.
> 
> These chunks will be used as the initial minimum allocation and increased
> in increments of the chunk using realloc.

Any reason to make it different by design than LFT allocations? I think
that similar to LFT case you always know how much memory is needed (how
many MLIDs are allocated) before multicast rerouting cycle.

Sasha

> 
> MFTs are only be increased in size and are never reduced in size. If a realloc
> for MFT fails, it is treated as a fatal error and OpenSM is exited.
> 
> Signed-off-by: Hal Rosenstock <hal.rosenstock-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> ---
> 
> diff --git a/opensm/include/opensm/osm_base.h b/opensm/include/opensm/osm_base.h
> index 06223ce..524646c 100644
> --- a/opensm/include/opensm/osm_base.h
> +++ b/opensm/include/opensm/osm_base.h
> @@ -449,6 +449,17 @@ BEGIN_C_DECLS
>  */
>  #define OSM_DEFAULT_SMP_MAX_ON_WIRE 4
>  /***********/
> +/****d* OpenSM: Base/OSM_DEFAULT_MFT_CHUNKS
> +* NAME
> +*	OSM_DEFAULT_MFT_CHUNKS
> +*
> +* DESCRIPTION
> +*	Specifies the default number of 64 entry chunks in MFT related
> +*	memory (re)allocation. Default is 16 (1K entries).
> +*
> +* SYNOPSIS
> +*/
> +#define OSM_DEFAULT_MFT_CHUNKS 16
>  /****d* OpenSM: Base/OSM_SM_DEFAULT_QP0_RCV_SIZE
>  * NAME
>  *	OSM_SM_DEFAULT_QP0_RCV_SIZE
> diff --git a/opensm/include/opensm/osm_mcast_tbl.h b/opensm/include/opensm/osm_mcast_tbl.h
> index 710d199..503a9cb 100644
> --- a/opensm/include/opensm/osm_mcast_tbl.h
> +++ b/opensm/include/opensm/osm_mcast_tbl.h
> @@ -1,6 +1,6 @@
>  /*
>   * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
> - * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
> + * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
>   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
>   *
>   * This software is available to you under a choice of one of two
> @@ -46,6 +46,7 @@
>  #include <iba/ib_types.h>
>  #include <complib/cl_qmap.h>
>  #include <opensm/osm_base.h>
> +#include <opensm/osm_subnet.h>
>  
>  #ifdef __cplusplus
>  #  define BEGIN_C_DECLS extern "C" {
> @@ -74,6 +75,7 @@ typedef struct osm_mcast_fwdbl {
>  	int16_t max_block_in_use;
>  	uint16_t num_entries;
>  	uint16_t max_mlid_ho;
> +	uint16_t mft_size;
>  	uint16_t(*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
>  } osm_mcast_tbl_t;
>  /*
> @@ -97,7 +99,7 @@ typedef struct osm_mcast_fwdbl {
>  *	max_mlid_ho
>  *		Maximum MLID value (host order).
>  *
> -*	pp_mask_tbl
> +*	p_mask_tbl
>  *		Pointer to a two dimensional array of port_masks for this switch.
>  *		The first dimension is MLID, the second dimension is mask position.
>  *		This pointer is null for switches that do not support multicast.
> @@ -115,7 +117,8 @@ typedef struct osm_mcast_fwdbl {
>  * SYNOPSIS
>  */
>  ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
> -				   IN uint8_t num_ports, IN uint16_t capacity);
> +				   IN uint8_t num_ports, IN uint16_t capacity,
> +				   IN osm_subn_t * const p_subn);
>  /*
>  * PARAMETERS
>  *	num_ports
> @@ -158,6 +161,39 @@ void osm_mcast_tbl_delete(IN osm_mcast_tbl_t ** pp_tbl);
>  * SEE ALSO
>  *********/
>  
> +/****f* OpenSM: Forwarding Table/osm_mcast_tbl_realloc_mask_tbl
> +* NAME
> +*	osm_mcast_tbl_realloc_mask_tbl
> +*
> +* DESCRIPTION
> +*	This function reallocates the port mask table if necessary.
> +*
> +* SYNOPSIS
> +*/
> +void
> +osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * const p_tbl,
> +			       IN osm_subn_t * const p_subn,
> +			       IN uintn_t mlid_offset);
> +/*
> +* PARAMETERS
> +*
> +*	p_tbl
> +*		[in] Pointer to the Multicast Forwarding Table object.
> +*
> +*	p_subn
> +*		[in] Pointer to the subnet object.
> +*
> +*	mlid_offset
> +*		[in] Offset of MLID being accessed.
> +*
> +* RETURN VALUE
> +*	None
> +*
> +* NOTES
> +*
> +* SEE ALSO
> +*/
> +
>  /****f* OpenSM: Forwarding Table/osm_mcast_tbl_destroy
>  * NAME
>  *	osm_mcast_tbl_destroy
> @@ -191,7 +227,7 @@ void osm_mcast_tbl_destroy(IN osm_mcast_tbl_t * p_tbl);
>  * SYNOPSIS
>  */
>  void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
> -		       IN uint8_t port_num);
> +		       IN uint8_t port_num, IN osm_subn_t * const p_subn);
>  /*
>  * PARAMETERS
>  *	p_tbl
> @@ -304,6 +340,7 @@ boolean_t osm_mcast_tbl_is_any_port(IN const osm_mcast_tbl_t * p_tbl,
>  * SYNOPSIS
>  */
>  ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
> +					IN osm_subn_t * p_subn,
>  					IN const ib_net16_t * p_block,
>  					IN int16_t block_num,
>  					IN uint8_t position);
> @@ -336,8 +373,8 @@ ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
>  * SYNOPSIS
>  */
>  boolean_t osm_mcast_tbl_get_block(IN osm_mcast_tbl_t * p_tbl,
> -				  IN int16_t block_num, IN uint8_t position,
> -				  OUT ib_net16_t * p_block);
> +				  IN osm_subn_t * p_subn, IN int16_t block_num,
> +				  IN uint8_t position, OUT ib_net16_t * p_block);
>  /*
>  * PARAMETERS
>  *	p_tbl
> diff --git a/opensm/include/opensm/osm_subnet.h b/opensm/include/opensm/osm_subnet.h
> index 9488225..2893d18 100644
> --- a/opensm/include/opensm/osm_subnet.h
> +++ b/opensm/include/opensm/osm_subnet.h
> @@ -211,6 +211,7 @@ typedef struct osm_subn_opt {
>  	osm_qos_options_t qos_rtr_options;
>  	boolean_t enable_quirks;
>  	boolean_t no_clients_rereg;
> +	uint32_t mft_chunks;
>  #ifdef ENABLE_OSM_PERF_MGR
>  	boolean_t perfmgr;
>  	boolean_t perfmgr_redir;
> @@ -428,6 +429,9 @@ typedef struct osm_subn_opt {
>  *	babbling_port_policy
>  *		OpenSM will enforce its "babbling" port policy.
>  *
> +*	mft_chunks
> +*		Number of 16 entry chunks used in MFT (re)allocation
> +*
>  *	perfmgr
>  *		Enable or disable the performance manager
>  *
> diff --git a/opensm/include/opensm/osm_switch.h b/opensm/include/opensm/osm_switch.h
> index 655491d..0262d10 100644
> --- a/opensm/include/opensm/osm_switch.h
> +++ b/opensm/include/opensm/osm_switch.h
> @@ -222,7 +222,8 @@ void osm_switch_delete(IN OUT osm_switch_t ** pp_sw);
>  * SYNOPSIS
>  */
>  osm_switch_t *osm_switch_new(IN osm_node_t * p_node,
> -			     IN const osm_madw_t * p_madw);
> +			     IN const osm_madw_t * p_madw,
> +			     IN osm_subn_t * p_subn);
>  /*
>  * PARAMETERS
>  *	p_node
> @@ -747,12 +748,13 @@ osm_switch_set_lft_block(IN osm_switch_t * p_sw, IN const uint8_t * p_block,
>  * SYNOPSIS
>  */
>  static inline ib_api_status_t
> -osm_switch_set_mft_block(IN osm_switch_t * p_sw, IN const ib_net16_t * p_block,
> +osm_switch_set_mft_block(IN osm_switch_t * p_sw, IN osm_subn_t * const p_subn,
> +			 IN const ib_net16_t * p_block,
>  			 IN uint16_t block_num, IN uint8_t position)
>  {
>  	CL_ASSERT(p_sw);
> -	return osm_mcast_tbl_set_block(&p_sw->mcast_tbl, p_block, block_num,
> -				       position);
> +	return osm_mcast_tbl_set_block(&p_sw->mcast_tbl, p_subn, p_block,
> +				       block_num, position);
>  }
>  /*
>  * PARAMETERS
> @@ -786,13 +788,14 @@ osm_switch_set_mft_block(IN osm_switch_t * p_sw, IN const ib_net16_t * p_block,
>  * SYNOPSIS
>  */
>  static inline boolean_t osm_switch_get_mft_block(IN osm_switch_t * p_sw,
> +						 IN osm_subn_t * const p_subn,
>  						 IN uint16_t block_num,
>  						 IN uint8_t position,
>  						 OUT ib_net16_t * p_block)
>  {
>  	CL_ASSERT(p_sw);
> -	return osm_mcast_tbl_get_block(&p_sw->mcast_tbl, block_num, position,
> -				       p_block);
> +	return osm_mcast_tbl_get_block(&p_sw->mcast_tbl, p_subn, block_num,
> +				       position, p_block);
>  }
>  /*
>  * PARAMETERS
> diff --git a/opensm/opensm/osm_dump.c b/opensm/opensm/osm_dump.c
> index 08b3156..c004b6c 100644
> --- a/opensm/opensm/osm_dump.c
> +++ b/opensm/opensm/osm_dump.c
> @@ -1,7 +1,7 @@
>  /*
>   * Copyright (c) 2009 Sun Microsystems, Inc. All rights reserved.
>   * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
> - * Copyright (c) 2002-2006 Mellanox Technologies LTD. All rights reserved.
> + * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
>   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
>   *
>   * This software is available to you under a choice of one of two
> @@ -232,6 +232,7 @@ static void dump_ucast_routes(cl_map_item_t * item, FILE * file, void *cxt)
>  static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt)
>  {
>  	osm_switch_t *p_sw = (osm_switch_t *) item;
> +	osm_opensm_t *p_osm = cxt;
>  	osm_mcast_tbl_t *p_tbl;
>  	int16_t mlid_ho = 0;
>  	int16_t mlid_start_ho;
> @@ -261,6 +262,9 @@ static void dump_mcast_routes(cl_map_item_t * item, FILE * file, void *cxt)
>  			sprintf(mlid_hdr, "0x%04X :",
>  				mlid_ho + IB_LID_MCAST_START_HO);
>  			while (position <= p_tbl->max_position) {
> +				osm_mcast_tbl_realloc_mask_tbl(p_tbl,
> +							       &p_osm->subn,
> +							       mlid_ho);
>  				mask_entry =
>  				    cl_ntoh16((*p_tbl->
>  					       p_mask_tbl)[mlid_ho][position]);
> diff --git a/opensm/opensm/osm_mcast_fwd_rcv.c b/opensm/opensm/osm_mcast_fwd_rcv.c
> index f3d0183..b071953 100644
> --- a/opensm/opensm/osm_mcast_fwd_rcv.c
> +++ b/opensm/opensm/osm_mcast_fwd_rcv.c
> @@ -103,7 +103,7 @@ void osm_mft_rcv_process(IN void *context, IN void *data)
>  			"MFT received for nonexistent node "
>  			"0x%016" PRIx64 "\n", cl_ntoh64(node_guid));
>  	} else {
> -		status = osm_switch_set_mft_block(p_sw, p_block,
> +		status = osm_switch_set_mft_block(p_sw, sm->p_subn, p_block,
>  						  (uint16_t) block_num,
>  						  position);
>  		if (status != IB_SUCCESS) {
> diff --git a/opensm/opensm/osm_mcast_mgr.c b/opensm/opensm/osm_mcast_mgr.c
> index 77e0b94..55369d0 100644
> --- a/opensm/opensm/osm_mcast_mgr.c
> +++ b/opensm/opensm/osm_mcast_mgr.c
> @@ -322,7 +322,7 @@ static int mcast_mgr_set_mft_block(osm_sm_t * sm, IN osm_switch_t * p_sw,
>  
>  	p_tbl = osm_switch_get_mcast_tbl_ptr(p_sw);
>  
> -	if (osm_mcast_tbl_get_block(p_tbl, (uint16_t) block_num,
> +	if (osm_mcast_tbl_get_block(p_tbl, sm->p_subn, (uint16_t) block_num,
>  				    (uint8_t) position, block)) {
>  
>  		block_id_ho = block_num + (position << 28);
> @@ -571,7 +571,7 @@ static osm_mtree_node_t *mcast_mgr_branch(osm_sm_t * sm, osm_mgrp_t * p_mgrp,
>  			"Adding upstream port %u\n", upstream_port);
>  
>  		CL_ASSERT(upstream_port);
> -		osm_mcast_tbl_set(p_tbl, mlid_ho, upstream_port);
> +		osm_mcast_tbl_set(p_tbl, mlid_ho, upstream_port, sm->p_subn);
>  	}
>  
>  	/*
> @@ -610,7 +610,7 @@ static osm_mtree_node_t *mcast_mgr_branch(osm_sm_t * sm, osm_mgrp_t * p_mgrp,
>  		   set the appropriate bit in the multicast forwarding
>  		   table for this switch.
>  		 */
> -		osm_mcast_tbl_set(p_tbl, mlid_ho, i);
> +		osm_mcast_tbl_set(p_tbl, mlid_ho, i, sm->p_subn);
>  		if (i == 0) {
>  			/* This means we are adding the switch to the MC group.
>  			   We do not need to continue looking at the remote port, just
> @@ -812,7 +812,7 @@ void osm_mcast_mgr_set_table(osm_sm_t * sm, IN const osm_mgrp_t * p_mgrp,
>  		if (p_child_mtn == NULL)
>  			continue;
>  
> -		osm_mcast_tbl_set(p_tbl, mlid_ho, i);
> +		osm_mcast_tbl_set(p_tbl, mlid_ho, i, sm->p_subn);
>  	}
>  
>  	OSM_LOG_EXIT(sm->p_log);
> @@ -941,7 +941,8 @@ ib_api_status_t osm_mcast_mgr_process_single(osm_sm_t * sm,
>  
>  			p_mcast_tbl =
>  			    osm_switch_get_mcast_tbl_ptr(p_remote_node->sw);
> -			osm_mcast_tbl_set(p_mcast_tbl, mlid_ho, port_num);
> +			osm_mcast_tbl_set(p_mcast_tbl, mlid_ho, port_num,
> +					  sm->p_subn);
>  		} else {
>  			if (join_state & IB_JOIN_STATE_SEND_ONLY)
>  				OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
> diff --git a/opensm/opensm/osm_mcast_tbl.c b/opensm/opensm/osm_mcast_tbl.c
> index d7c9529..4cc775b 100644
> --- a/opensm/opensm/osm_mcast_tbl.c
> +++ b/opensm/opensm/osm_mcast_tbl.c
> @@ -1,6 +1,6 @@
>  /*
>   * Copyright (c) 2004-2006 Voltaire, Inc. All rights reserved.
> - * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
> + * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
>   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
>   * Copyright (c) 2009 HNR Consulting. All rights reserved.
>   *
> @@ -50,11 +50,14 @@
>  #include <complib/cl_math.h>
>  #include <iba/ib_types.h>
>  #include <opensm/osm_mcast_tbl.h>
> +#include <opensm/osm_log.h>
> +#include <opensm/osm_opensm.h>
>  
>  /**********************************************************************
>   **********************************************************************/
>  ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
> -				   IN uint8_t num_ports, IN uint16_t capacity)
> +				   IN uint8_t num_ports, IN uint16_t capacity,
> +				   IN osm_subn_t * p_subn)
>  {
>  	CL_ASSERT(p_tbl);
>  	CL_ASSERT(num_ports);
> @@ -93,7 +96,8 @@ ib_api_status_t osm_mcast_tbl_init(IN osm_mcast_tbl_t * p_tbl,
>  	   since it is (and must be) defined that way the table structure
>  	   in order to create a pointer to a two dimensional array.
>  	 */
> -	p_tbl->p_mask_tbl = calloc(p_tbl->num_entries,
> +	p_tbl->mft_size = p_subn->opt.mft_chunks * IB_MCAST_BLOCK_SIZE;
> +	p_tbl->p_mask_tbl = calloc(p_tbl->mft_size,
>  				   (IB_MCAST_POSITION_MAX +
>  				    1) * IB_MCAST_MASK_SIZE / 8);
>  
> @@ -113,7 +117,7 @@ void osm_mcast_tbl_destroy(IN osm_mcast_tbl_t * p_tbl)
>  /**********************************************************************
>   **********************************************************************/
>  void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
> -		       IN uint8_t port)
> +		       IN uint8_t port, IN osm_subn_t * p_subn)
>  {
>  	uintn_t mlid_offset;
>  	uintn_t mask_offset;
> @@ -128,6 +132,7 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
>  	mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
>  	mask_offset = port / IB_MCAST_MASK_SIZE;
>  	bit_mask = cl_ntoh16((uint16_t) (1 << (port % IB_MCAST_MASK_SIZE)));
> +	osm_mcast_tbl_realloc_mask_tbl(p_tbl, p_subn, mlid_offset);
>  	(*p_tbl->p_mask_tbl)[mlid_offset][mask_offset] |= bit_mask;
>  
>  	block_num = (int16_t) (mlid_offset / IB_MCAST_BLOCK_SIZE);
> @@ -138,6 +143,45 @@ void osm_mcast_tbl_set(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho,
>  
>  /**********************************************************************
>   **********************************************************************/
> +void
> +osm_mcast_tbl_realloc_mask_tbl(IN osm_mcast_tbl_t * const p_tbl,
> +			       IN osm_subn_t * const p_subn,
> +			       IN uintn_t mlid_offset)
> +{
> +	size_t mft_size, size;
> +	uint16_t (*p_mask_tbl)[][IB_MCAST_POSITION_MAX];
> +
> +	if (mlid_offset < p_tbl->mft_size)
> +		return;
> +
> +	mft_size = (mlid_offset +
> +		    p_subn->opt.mft_chunks * IB_MCAST_BLOCK_SIZE) /
> +		    IB_MCAST_BLOCK_SIZE * IB_MCAST_BLOCK_SIZE;
> +	if (mft_size == p_tbl->mft_size)
> +		mft_size += p_subn->opt.mft_chunks * IB_MCAST_BLOCK_SIZE;
> +	if (mft_size > p_tbl->max_block * IB_MCAST_BLOCK_SIZE)
> +		mft_size = p_tbl->max_block * IB_MCAST_BLOCK_SIZE;
> +	size = mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8;
> +	p_mask_tbl = realloc(p_tbl->p_mask_tbl, size);
> +	if (!p_mask_tbl)
> +		goto error;
> +	memset((uint8_t *)p_mask_tbl + p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8,
> +	       0,
> +	       size - p_tbl->mft_size * (IB_MCAST_POSITION_MAX + 1) * IB_MCAST_MASK_SIZE / 8);
> +	p_tbl->p_mask_tbl = p_mask_tbl;
> +	p_tbl->mft_size = mft_size;
> +	return;
> +
> +error:
> +	OSM_LOG(&p_subn->p_osm->log, OSM_LOG_SYS,
> +		"Reallocation of multicast mask table failed - exiting\n");
> +	OSM_LOG(&p_subn->p_osm->log, OSM_LOG_ERROR, " ERR 6401: "
> +		"Reallocation of multicast mask table failed - exiting\n");
> +	exit(1);
> +}
> +
> +/**********************************************************************
> + **********************************************************************/
>  boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
>  				IN uint16_t mlid_ho, IN uint8_t port_num)
>  {
> @@ -154,6 +198,8 @@ boolean_t osm_mcast_tbl_is_port(IN const osm_mcast_tbl_t * p_tbl,
>  		CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
>  
>  		mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
> +		if (mlid_offset >= p_tbl->mft_size)
> +			return FALSE;
>  		mask_offset = port_num / IB_MCAST_MASK_SIZE;
>  		bit_mask = cl_ntoh16((uint16_t)
>  				     (1 << (port_num % IB_MCAST_MASK_SIZE)));
> @@ -181,6 +227,8 @@ boolean_t osm_mcast_tbl_is_any_port(IN const osm_mcast_tbl_t * p_tbl,
>  		CL_ASSERT(mlid_ho <= p_tbl->max_mlid_ho);
>  
>  		mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
> +		if (mlid_offset >= p_tbl->mft_size)
> +			return FALSE;
>  
>  		for (position = 0; position <= p_tbl->max_position; position++)
>  			result |= (*p_tbl->p_mask_tbl)[mlid_offset][position];
> @@ -192,6 +240,7 @@ boolean_t osm_mcast_tbl_is_any_port(IN const osm_mcast_tbl_t * p_tbl,
>  /**********************************************************************
>   **********************************************************************/
>  ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
> +					IN osm_subn_t * p_subn,
>  					IN const ib_net16_t * p_block,
>  					IN int16_t block_num,
>  					IN uint8_t position)
> @@ -213,6 +262,9 @@ ib_api_status_t osm_mcast_tbl_set_block(IN osm_mcast_tbl_t * p_tbl,
>  	if (mlid_start_ho + IB_MCAST_BLOCK_SIZE - 1 > p_tbl->max_mlid_ho)
>  		return IB_INVALID_PARAMETER;
>  
> +	osm_mcast_tbl_realloc_mask_tbl(p_tbl, p_subn,
> +				       mlid_start_ho + IB_MCAST_BLOCK_SIZE);
> +
>  	for (i = 0; i < IB_MCAST_BLOCK_SIZE; i++)
>  		(*p_tbl->p_mask_tbl)[mlid_start_ho + i][position] = p_block[i];
>  
> @@ -234,6 +286,8 @@ void osm_mcast_tbl_clear_mlid(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho)
>  
>  	if (p_tbl->p_mask_tbl && (mlid_ho <= p_tbl->max_mlid_ho)) {
>  		mlid_offset = mlid_ho - IB_LID_MCAST_START_HO;
> +		if (mlid_offset >= p_tbl->mft_size)
> +			return;
>  		for (i = 0; i <= p_tbl->max_position; i++)
>  			(*p_tbl->p_mask_tbl)[mlid_offset][i] = 0;
>  	}
> @@ -242,6 +296,7 @@ void osm_mcast_tbl_clear_mlid(IN osm_mcast_tbl_t * p_tbl, IN uint16_t mlid_ho)
>  /**********************************************************************
>   **********************************************************************/
>  boolean_t osm_mcast_tbl_get_block(IN osm_mcast_tbl_t * p_tbl,
> +				  IN osm_subn_t * p_subn,
>  				  IN int16_t block_num, IN uint8_t position,
>  				  OUT ib_net16_t * p_block)
>  {
> @@ -264,6 +319,9 @@ boolean_t osm_mcast_tbl_get_block(IN osm_mcast_tbl_t * p_tbl,
>  
>  	mlid_start_ho = (uint16_t) (block_num * IB_MCAST_BLOCK_SIZE);
>  
> +	osm_mcast_tbl_realloc_mask_tbl(p_tbl, p_subn,
> +				       mlid_start_ho + IB_MCAST_BLOCK_SIZE);
> +
>  	for (i = 0; i < IB_MCAST_BLOCK_SIZE; i++)
>  		p_block[i] = (*p_tbl->p_mask_tbl)[mlid_start_ho + i][position];
>  
> diff --git a/opensm/opensm/osm_sa_mft_record.c b/opensm/opensm/osm_sa_mft_record.c
> index 841eb86..79a4f44 100644
> --- a/opensm/opensm/osm_sa_mft_record.c
> +++ b/opensm/opensm/osm_sa_mft_record.c
> @@ -1,6 +1,6 @@
>  /*
>   * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
> - * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
> + * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
>   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
>   *
>   * This software is available to you under a choice of one of two
> @@ -104,7 +104,8 @@ static ib_api_status_t mftr_rcv_new_mftr(IN osm_sa_t * sa,
>  	p_rec_item->rec.position_block_num = cl_hton16(position_block_num);
>  
>  	/* copy the mft block */
> -	osm_switch_get_mft_block(p_sw, block, position, p_rec_item->rec.mft);
> +	osm_switch_get_mft_block(p_sw, sa->p_subn, block, position,
> +				 p_rec_item->rec.mft);
>  
>  	cl_qlist_insert_tail(p_list, &p_rec_item->list_item);
>  
> diff --git a/opensm/opensm/osm_subnet.c b/opensm/opensm/osm_subnet.c
> index 647950e..bcaea8a 100644
> --- a/opensm/opensm/osm_subnet.c
> +++ b/opensm/opensm/osm_subnet.c
> @@ -352,6 +352,7 @@ static const opt_rec_t opt_tbl[] = {
>  	{ "daemon", OPT_OFFSET(daemon), opts_parse_boolean, NULL, 0 },
>  	{ "sm_inactive", OPT_OFFSET(sm_inactive), opts_parse_boolean, NULL, 1 },
>  	{ "babbling_port_policy", OPT_OFFSET(babbling_port_policy), opts_parse_boolean, NULL, 1 },
> +	{ "mft_chunks", OPT_OFFSET(mft_chunks), opts_parse_uint32, NULL, 1 },
>  #ifdef ENABLE_OSM_PERF_MGR
>  	{ "perfmgr", OPT_OFFSET(perfmgr), opts_parse_boolean, NULL, 0 },
>  	{ "perfmgr_redir", OPT_OFFSET(perfmgr_redir), opts_parse_boolean, NULL, 0 },
> @@ -724,6 +725,7 @@ void osm_subn_set_default_opt(IN osm_subn_opt_t * p_opt)
>  	p_opt->daemon = FALSE;
>  	p_opt->sm_inactive = FALSE;
>  	p_opt->babbling_port_policy = FALSE;
> +	p_opt->mft_chunks = OSM_DEFAULT_MFT_CHUNKS;
>  #ifdef ENABLE_OSM_PERF_MGR
>  	p_opt->perfmgr = FALSE;
>  	p_opt->perfmgr_redir = TRUE;
> @@ -1199,6 +1201,13 @@ int osm_subn_parse_conf_file(char *file_name, osm_subn_opt_t * p_opts)
>  				    NULL);
>  			break;
>  		}
> +		if (p_opts->mft_chunks < 1 || p_opts->mft_chunks > 256) {
> +			log_report(" Invalid Cached Option Value:"
> +				   "mft_chunks = %u"
> +				   " Using Default:%u\n",
> +				   p_opts->mft_chunks, OSM_DEFAULT_MFT_CHUNKS);
> +			p_opts->mft_chunks = OSM_DEFAULT_MFT_CHUNKS;
> +		}
>  	}
>  	fclose(opts_file);
>  
> @@ -1524,6 +1533,11 @@ int osm_subn_output_conf(FILE *out, IN osm_subn_opt_t * p_opts)
>  		p_opts->sm_inactive ? "TRUE" : "FALSE",
>  		p_opts->babbling_port_policy ? "TRUE" : "FALSE");
>  
> +	fprintf(out,
> +		"# Number of 16 entry chunks used when (re)allocating "
> +		"MFTs\nmft_chunks %d\n\n",
> +		p_opts->mft_chunks);
> +
>  #ifdef ENABLE_OSM_PERF_MGR
>  	fprintf(out,
>  		"#\n# Performance Manager Options\n#\n"
> diff --git a/opensm/opensm/osm_sw_info_rcv.c b/opensm/opensm/osm_sw_info_rcv.c
> index c335263..9861525 100644
> --- a/opensm/opensm/osm_sw_info_rcv.c
> +++ b/opensm/opensm/osm_sw_info_rcv.c
> @@ -1,6 +1,6 @@
>  /*
>   * Copyright (c) 2004-2008 Voltaire, Inc. All rights reserved.
> - * Copyright (c) 2002-2005 Mellanox Technologies LTD. All rights reserved.
> + * Copyright (c) 2002-2009 Mellanox Technologies LTD. All rights reserved.
>   * Copyright (c) 1996-2003 Intel Corporation. All rights reserved.
>   *
>   * This software is available to you under a choice of one of two
> @@ -211,7 +211,7 @@ static void si_rcv_process_new(IN osm_sm_t * sm, IN osm_node_t * p_node,
>  
>  	osm_dump_switch_info(sm->p_log, p_si, OSM_LOG_DEBUG);
>  
> -	p_sw = osm_switch_new(p_node, p_madw);
> +	p_sw = osm_switch_new(p_node, p_madw, sm->p_subn);
>  	if (p_sw == NULL) {
>  		OSM_LOG(sm->p_log, OSM_LOG_ERROR, "ERR 3608: "
>  			"Unable to allocate new switch object\n");
> diff --git a/opensm/opensm/osm_switch.c b/opensm/opensm/osm_switch.c
> index ed0bc66..7ad1af4 100644
> --- a/opensm/opensm/osm_switch.c
> +++ b/opensm/opensm/osm_switch.c
> @@ -99,7 +99,8 @@ void osm_switch_delete(IN OUT osm_switch_t ** pp_sw)
>  /**********************************************************************
>   **********************************************************************/
>  osm_switch_t *osm_switch_new(IN osm_node_t * p_node,
> -			     IN const osm_madw_t * p_madw)
> +			     IN const osm_madw_t * p_madw,
> +			     IN osm_subn_t * p_subn)
>  {
>  	osm_switch_t *p_sw;
>  	ib_switch_info_t *p_si;
> @@ -137,7 +138,7 @@ osm_switch_t *osm_switch_new(IN osm_node_t * p_node,
>  	memset(p_sw->p_prof, 0, sizeof(*p_sw->p_prof) * num_ports);
>  
>  	if (osm_mcast_tbl_init(&p_sw->mcast_tbl, osm_node_get_num_physp(p_node),
> -			       cl_ntoh16(p_si->mcast_cap)))
> +			       cl_ntoh16(p_si->mcast_cap), p_subn))
>  		goto err;
>  
>  	for (port_num = 0; port_num < num_ports; port_num++)
> @@ -508,7 +509,6 @@ static int alloc_lft(IN osm_switch_t * p_sw, uint16_t lids)
>  		p_sw->lft = new_lft;
>  		p_sw->lft_size = lft_size;
>  	}
> -
>  	return 0;
>  }
>  
> @@ -549,7 +549,6 @@ int osm_switch_prepare_path_rebuild(IN osm_switch_t * p_sw, IN uint16_t max_lids
>  		p_sw->num_hops = max_lids + 1;
>  	}
>  	p_sw->max_lid_ho = max_lids;
> -
>  	return 0;
>  }
>  
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html